DNA methyltransferase 3A promotes cell proliferation by silencing CDK inhibitor p18INK4C in gastric carcinogenesis

Little is known about the roles of DNA methyltransferase 3A (DNMT3A) in gastric carcinogenesis. Here, we reported that the exogenous expression of DNMT3A promoted gastric cancer (GC) cell proliferation by accelerating the G1/S transition. Subsequently, p18INK4C was identified as a downstream target of DNMT3A. The elevated expression of DNMT3A suppressed p18INK4C at least at the transcriptional level. Depletion of p18INK4C expression in GC cells induced cell cycle progression, whereas its re-expression alleviated the effect of DNMT3A overexpression on G1/S transition. Furthermore, we found that DNMT3A modulated p18INK4C by directly binding to and silencing the p18INK4C gene via promoter hypermethylation. In clinical GC tissue specimens analyzed, the level of methylation of p18INK4C detected in tumor tissues was significantly higher than that in paired non-tumor tissues. Moreover, elevated level of DNMT3A expression was associated with the differentiation of GC tissues and was negatively correlated with the p18INK4C expression level. Taken together, our results found that DNMT3A contributes to the dysregulation of the cell cycle by repressing p18INK4C in a DNA methylation-dependent manner, suggesting that DNMT3A-p18INK4C axis involved in GC. These findings provide new insights into gastric carcinogenesis and a potential therapeutic target for GC that may be further investigated in the future.

DNMT3A may be more critical for GC progression than that of the other two DNMTs. Many studies have shown that abnormal DNA methylation in GC alters the expression of tumor suppressor genes (TSGs) [14][15][16][17] . Therefore, further investigation of DNMT3A is needed to explore the precise role or mechanism underlying the regulation of GC.
In the gastrointestinal epithelium, cell proliferation and differentiation are highly regulated processes governed by intrinsic factors, such as cell cycle regulators 18 . Previous studies have demonstrated that inhibitors of CDK4 (INK4)-CDK4/6-CyclinD-Rb-E2F pathway play a key role in controlling cell growth 19 . The INK4 family includes p16 INK4A , p15 INK4B , p18 INK4C , and p19 INK4D , and its inactivation can lead to the formation of active CDK4/6-CyclinD complexes and further promote cell cycle progression 20 . In GC, the de-regulation of p16 has been shown to significantly increase the risk of malignant transformation of gastric epithelial cells 21 , and the silencing of INK4 members induced by Ras homolog family member A (RhoA) has been associated with G1/S progression, indicating that INK4 members are involved in GC cell proliferation 22 . In addition, the silencing of INK4 members via promoter hypermethylation has been shown to occur in certain cancers [23][24][25] . However, it remains unclear whether the increased expression of DNMT3A in GC accounts for the dysregulation of INK4 members.
In this study, we investigated the expression pattern and biological function of DNMT3A in GC as well as DNA methylation mechanism resulting from its activity. We have shown that DNMT3A is involved in GC progression via methylation of the p18 INK4C promoter, which leads to the downregulation of p18 INK4C , thereby disrupting the G1/S checkpoint and eventually promoting GC cell proliferation. These findings may be beneficial to the development of new treatment options for GC that target DNMT3A.

DNMT3A is important for GC cell proliferation. Abnormal cell proliferation is a characteristic
feature of cells that have undergone malignant transformation. DNMT3A has been implicated in cell survival in melanoma and hepatocellular carcinoma 26,27 . To evaluate the functional outcomes of DNMT3A in GC progression, a cell model for DNMT3A analysis was generated. AGS and BGC-823 cells were selected to establish stable DNMT3A knockdown GC cell lines (named AGS-shDNTM3A and BGC-shDNTM3A). Compared with control cells (named AGS-shControl and BGC-shControl), DNMT3A protein expression was dramatically decreased in AGS-shDNTM3A and BGC-shDNTM3A cells ( Figure  S1a). The biological roles of DNMT3A were then assessed via cell growth rate and foci formation assays in vitro. The results showed that DNMT3A knockdown AGS and BGC-823 cells had significantly reduced growth rates (P < 0.01; Fig. 1a) and lower frequencies of foci formation (P < 0.01; Fig. 1b) compared with control cells. Conversely, we examined the effect of the ectopic expression of DNMT3A on the growth of MKN45 cells. Exogenous DNMT3A expression in stably transfected MKN45 cells (named MKN45-DNMT3A) was confirmed by western blot analysis ( Figure S1b). DNMT3A overexpression caused a significant increase in the cell growth rate (P < 0.05; Figure S2a), suggesting that DNMT3A is required for GC cell proliferation. To further investigate the contribution of DNMT3A to cell proliferation, we investigated its role in nude mice in vivo. The tumor growth curves of BGC-shControl and BGC-shDNMT3A cells in nude mice are shown in Fig. 1c. The tumor volume was significantly smaller in BGC-shDNMT3A cell-injected nude mice (7.5 ± 9.23 mm 3 ) than in BGC-shControl cell-injected mice (159.5 ± 151.26 mm 3 ; P < 0.05; Fig. 1c and Figure S2b), indicating that DNMT3A has a critical function in tumor growth. Collectively, these results demonstrate that DNMT3A promotes tumor cell proliferation in vitro and in vivo and therefore may contribute to maintaining malignant phenotype in GC. DNMT3A promotes cell growth by disrupting the G1/S checkpoint. De-regulation of cell growth in tumor cells is usually associated with the concomitant acceleration of cell cycle progression. We therefore examined the contribution of the cell cycle to the observed growth rate induced by DNMT3A. AGS-shControl and AGS-shDNMT3A cells were synchronized, with most cells in the G1 phase (72%). After being released from synchronized for 12 h, the percentage of cells in the G1 phase was significantly higher for AGS-shDNMT3A cells than for AGS-shControl cells (57.16% versus 41.93%; P < 0.05; Fig. 1d). Reciprocally, we synchronized BGC-823 cells stably transfected with a DNMT3A construct (named BGC-DNMT3A cells) or empty vector (named BGC-Control cells) (Figure S1b), with most cells in the G1 phase (70%). After being released from synchronized for 12 h, the percentage of cells in the G1 phase was significantly lower for BGC-DNMT3A cells than for BGC-Control cells (21.52% versus 42.30%; P < 0.05; Fig. 1e). This effect was also observed in untreated MKN45-DNMT3A cells, which showed a significantly decreased percentage of cells in the G1 phase compared with MKN45-Control cells (53.60% versus 66.21%; P < 0.05; Fig. 1f). To elucidate the molecular basis of the acceleration of the G1/S transition, we examined several G1/S regulators in DNMT3A overexpression cells or knockdown cells. The results showed that the levels of cyclinD1, CDK4 and CDK6 were increased in DNMT3A-induced MKN45 and BGC-823 cells. Conversely, inhibition of DNMT3A reduced the protein levels of cyclinD1, CDK4 and CDK6 (Fig. 1g), suggesting that DNMT3A stimulated cell cycle progression by disrupting the G1/S checkpoint.
Identification of target genes modulated by DNMT3A. The finding of a critical function of DNMT3A in the cell cycle motivated us to identify its target genes associated with the G1/S transition. Differential expression profiles between control and MKN45-DNMT3A cells were examined via complementary DNA (cDNA) microarray. Results showed that 906 genes expression level were remarkably altered (fold-change cutoff: 2.0). Of these altered genes, 558 genes were downregulated and 348 genes were upregulated. Gene ontology (GO) analysis of the dysregulated genes indicated that the predominant molecular functions were binding and catalytic activity (Fig. 2a). In the biological processes category, the top ten largest groups of dysregulated genes in response to DNMT3A overexpression were assigned to the GO terms cell proliferation, cell adhesion, cell cycle process and other processes (Fig. 2b), indicating that DNMT3A plays an oncogenic role in these processes.
According to the above analysis, 5 downregulated genes related to cell cycle were found, including E2F2, RBL1, RHOU, PRIM1 and BLM. To better gain insights into the molecular mechanism of the G1/S transition by DNMT3A, we added 22 downregulated genes as candidate by adjusting fold-change cutoff of the microarray. GO analysis revealed that 5 of 27 genes, including BLM, CDKN1B, CCNE2, p18 INK4C and CKS1B, were associated with the regulation of CDK activity. Of the 5 genes fulfilling this criterion, CDKN1B and p18 INK4C have been previously implicated in processes relevant to the inhibition of cyclinD1 and CDK4/6 complex formation during the G1 phase 20 . However, knockdown of DNMT3A expression did not have a remarkable effect on the expression of CDKN1B at the protein levels ( Figure  S3a), suggesting that p18 INK4C may be a target of DNMT3A-mediated dysregulation of cell cycle. To further validate the microarray results for p18 INK4C , quantitative real-time PCR (qPCR) was used to investigate the effect of DNMT3A on p18 INK4C expression, along with the other INK4 members, including p15 INK4B , p16 INK4A and p19 INK4D . The results showed that the depletion of DNMT3A increased the transcriptional levels of the four members of the INK4 family (Fig. 2c). Conversely, the suppression of the INK4 members was observed in DNMT3A-induced GC cells (Fig. 2d). Notably, p18 INK4C expression was more significantly changed than that of the other INK4 members. We further found that the effect of DNMT3A on p18 INK4C inhibition could be reversed by treatment of MKN45-DNMT3A cells with the DNA methyltransferase inhibitor 5-azacytidine (5-Aza) (Fig. 2e), suggesting that the p18 INK4C gene is the major downstream target of DNMT3A. DNMT1 and DNMT3B have been associated with TSGs silencing in human tumors. To confirm that the suppression of p18 INK4C is specific to DNMT3A overexpression, the mRNA levels of INK4 members were detected following the silencing of DNMT3B or DNMT1 in the AGS cell line by RNAi using siRNA against DNMT3B or DNMT1. The results showed that the expression of p19 INK4D was increased in DNMT3B knockdown cells (defined as a greater than 2-fold increase), whereas no significant changes were observed in the mRNA levels of the other INK4 members ( Figure S4a). In addition, the depletion of DNMT1 increased the transcriptional levels of four members of the INK4 family ( Figure S4b). However, the increases in the expression levels of p16 INK4A , p15 INK4B and p19 INK4D were higher than that of p18 INK4C in DNMT1 knockdown cells. Although the expression of p18 INK4C was partially affected by DNMT1, no significant upregulation of DNMT1 (P > 0.05, Figure S4c) or DNMT3B (P > 0.05, Figure  S4d) mRNA levels was observed in 35 GC tissues (the same tissues used to detect DNMT3A expression) compared with non-tumor tissues. Likewise, we did not find significant increases in DNMT3B or DNMT1 expression in manipulated AGS, BGC-823 or MKN45 cells used in this study (P > 0.05, Figure  S4e). Collectively, these data indicate that the suppressive effect of p18 INK4C is due to DNMT3A and not to DNMT1 or DNMT3B. p18 INK4C is involved in DNMT3A-mediated G1/S transition. To confirm that the DNMT3A-p18 INK4C axis contributes to cell cycle regulation, we tested whether p18 INK4C arrests GC cells at the G1/S phase. Flow cytometry analysis revealed that depletion of p18 INK4C with a specific siRNA (named sip18 INK4C cells) increased the G1/S transition compared with a negative control siRNA (named siControl cells) (P < 0.05; Fig. 2f). Western blot analysis showed significant increases in the protein levels of CDK4 and CDK6 in p18 INK4C knockdown cells (Fig. 2f). Subsequently, we investigated whether enforced p18 INK4C expression in DNMT3A overexpression cells could reduce cell growth. Compared with control cells, both MKN45-DNMT3A and BGC-DNMT3A cells showed significantly reduced cell growth rates (P < 0.01; Fig. 2g) and restoration of the percentages of cells in the G1 phase (P < 0.01; Fig. 2h) after re-expression of p18 INK4C . Moreover, the protein levels of CDK4 and CDK6 were found to be repressed in dose-dependent manners by the ectopic expression of p18 INK4C (Fig. 2i), suggesting that the restoration of p18 INK4C expression effectively prevented CDK4 and CDK6 activation and delayed the cell cycle progression mediated by DNMT3A overexpression. Collectively, these data indicated that p18 INK4C participated in the DNMT3A-induced G1/S transition.

Repression of p18 INK4C via DNMT3A-mediated DNA methylation. To evaluate whether
DNA methylation is involved in DNMT3A-induced p18 INK4C silencing, bisulfite DNA sequencing (BGS) assay was performed to examine the methylation status of the promoter region of p18 INK4C in DNMT3A-depleted BGC cells. As shown in Fig. 3a, 23 individual CpG sites within CpG island regions (from − 199 to + 50 bp) were sequenced to identify methylated cytosine residues. The results showed that the frequency of p18 INK4C promoter methylation in BGC-shDNMT3A cells was 26%, which was significantly lower than the level measured in control cells (54%; defined as a greater than 2-fold decrease) (Fig. 3b). Furthermore, BGS analysis was performed to evaluate the other three INK4 members (Fig. 3c). The promoters of these members did not show marked decreases in the numbers of methylated CpG sites, indicating that DNMT3A-mediated promoter hypermethylation mainly inactivates p18 INK4C among Scientific RepoRts | 5:13781 | DOi: 10.1038/srep13781   Fig. 4a). Moreover, although each site displayed extensive methylation in tumor and adjacent non-tumor tissues, the methylation levels of 19 sites within the p18 INK4C promoter CpG island were higher in tumor tissues than in non-tumor tissues, especially CpG 7 and 9 (defined as a greater than 2-fold increase; Fig. 4b). Furthermore, we detected the mRNA expression of p18 INK4C in the above cases (Fig. 4c). Then, the GC patients were divided into two subgroups based on their relative p18 INK4C mRNA status. Interestingly, in the clinical cases with low p18 INK4C expression (defined as a less than 2-fold decrease in GC tissues compared with adjacent non-tumor tissues), the p18 INK4C promoter showed significantly higher methylation levels in tumor tissues than in adjacent non-tumor tissues, indicating a negative correlation between DNA methylation and p18 INK4C expression (P < 0.05; Fig. 4d). Thus, the methylation-associated silencing of p18 INK4C is involved in GC progression, suggesting that the DNMT3A-p18 INK4C axis may be an important regulator of gastric carcinogenesis.  western blot data (Fig. 5a,b). The average expression level of DNMT3A in the tumor tissues was significantly higher than that in the paired adjacent non-tumor tissues (0.68 versus 0.39; P < 0.05; Fig. 5c). The correlation between DNMT3A protein expression and GC clinical pathological features was then analyzed. A total of 35 patients were categorized into two groups according to the DNMT3A protein expression level. The patients with DNMT3A overexpression (defined as a greater than 2-fold increase) were assigned to the N < T group, while the remainder were classified into the N ≥ T group. The results revealed that DNMT3A overexpression was strongly correlated with tumor cell differentiation (Table 1). These data indicate that the elevated expression of DNMT3A is important for gastric carcinogenesis. Furthermore, we explored the relationship between DNMT3A protein expression and p18 INK4C mRNA expression. A total of 35 pairs of GC cases and adjacent non-tumor controls were used for the detection of p18 INK4C by qPCR. The results showed that the frequency of p18 INK4C downregulation (defined as a greater than 2-fold decrease) was 60% (21/35; Fig. 5d) and that its average expression in tumor tissues was significantly lower than that in paired adjacent non-tumor tissues (1.56 versus 3.05; P < 0.05; Fig. 5e). In addition, an association study showed that p18 INK4C mRNA expression was negatively correlated with DNMT3A protein expression in the 35 pairs of GC cases (R = − 0.605; P < 0.01; Fig. 5f).

Discussion
GC is one of the most frequently occurring, fatal malignancies worldwide (ranking 4 th for males and 5 th for females), and the prognosis of this disease remains very poor 6 . Thus, a better understanding of the molecular mechanisms involved in GC progression is urgently required. Gene silencing via the hypermethylation of promoter CpG islands is a common event in various human cancers 28 . Previous studies have shown that the suppression of DNMT3A dramatically inhibits melanoma growth via de-regulation of cell cycle-related genes in mouse melanoma 26 . Similarly, DNMT3A-mediated promoter methylation causes PTEN gene silencing and promotes hepatocellular carcinoma cell proliferation 27 . In addition, elevated DNMT3a activity promoted polyposis in Apc Min mice 29 . All of these results indicate that DNMT3A facilitates tumor development. In GC, DNMT3A is enriched in tumor tissues, and it is recognized as an independent poor prognostic factor 12,13 ; thus, it is important to assess its biological function and the molecular mechanism underlying its promotion of GC individually. In this study, cell growth curve, colony-forming and in vivo tumorigenicity assays were employed to assess the characteristics of GC cells induced by DNMT3A. The results revealed that DNMT3A suppression in GC cell lines decreased cell proliferation, inhibited colony formation and suppressed tumorigenicity in nude mice. Furthermore, we found that the effects of DNMT3A overexpression were related to dysregulation of the G1/S transition by upregulation of CDK4 and CDK6 expression. DNMT3A was found to have tumor-promoting effects both in vitro and in vivo, and these results were concordant with those of our subsequent GO analysis of gene expression profiles based on microarray data, which revealed the top ten most significantly enriched categories, including cell proliferation, cell adhesion, cell cycle process and other processes.
Previous studies have demonstrated that several target genes, such as PTEN and insulin-like growth factor-3, are regulated by DNMT3A in hepatocellular carcinoma 27,30 . However, the target specificity of DNMT3A in GC remains unclear. In this study, we found that DNMT3A altered the expression of the cell cycle-related molecule p18 INK4C by microarray analysis. Further validation revealed that the increased expression of p18 INK4C in DNMT3A-depleted GC cells was significantly higher than that of three other INK4 members, implying that p18 INK4C may be a major downstream gene involved in the DNMT3A-induced G1/S transition. The most critical point in cell cycle regulation is the G1/S checkpoint, which is controlled by cyclins, CDKs, and CDK-inhibitors (CKIs) 31 . Members of the INK4 family of CKIs function as negative regulators of kinase activities 32 . p18 INK4C , a member of the INK4 family, functions as a tumor suppressor in mice, and its deletion has been associated with the development of tumors when combined with other oncogenic stimuli [33][34][35] . Similarly, the finding of decreased p18 INK4C gene expression at high frequencies in human cancers further supports its negative regulatory role during tumorigenesis [36][37][38] . Here, we found that p18 INK4C was required for the G1/S transition and that its re-expression reduced cell growth and delayed cell cycle progression from the G1 to S phase by inactivating CDK4 and CDK6 in DNMT3A-overexpressing cells, indicating that a functional connection exists between p18 INK4C and DNMT3A.
All INK4 members contain a CpG island in their promoter, suggesting that DNA methylation may contribute to the silencing of these genes. Previous studies have shown that a point mutation or deletion rarely inactivates the p18 INK4C locus 39 , suggesting that one of the mechanisms of p18 INK4C inactivation is epigenetic silencing. Indeed, promoter hypermethylation of p18 INK4C has been reported in certain cancers [23][24][25] . Thus, we assessed the methylation statuses of INK4 members. Interestingly, the methylation   45 . In this study, we did not find any significant changes in the predicted Sp1 binding sites in DNMT3A-depleted cells or control cells, indicating that aberrant methylation caused p18 INK4C silencing by preventing the binding of other transcription factors. We next examined the expression pattern and methylation status of p18 INK4C in clinical GC tissue specimens. The percentage of methylated CpGs located within p18 INK4C was higher in tumor tissues than in paired adjacent non-tumor tissues. These data are consistent with our findings in GC cells, indicating that DNA methylation is a key regulatory mechanism of p18 INK4C silencing. In addition, we assessed DNMT3A expression by western blot assay, which differs from the previous detection method using immunohistochemistry. The results showed that DNMT3A overexpression was significantly positively correlated with cell differentiation in GC and that it was negatively correlated with decreased p18 INK4C levels. In summary, our study has revealed novel functions of DNMT3A during gastric carcinogenesis, in which the de-regulation of the DNMT3A-p18 INK4C axis plays a critical role in cell cycle progression. Cell cycle analysis. Approximately 1 × 10 6 cells were trypsinized, washed twice with phosphate-buffered saline containing 1% fetal bovine serum and incubated in phosphate-buffered saline containing 0.02% TritonX-100, 0.1 mg/ml RNase (Sigma-Aldrich) and 10 μ g/ml propidium iodide (Sigma-Aldrich) for 30 min at 37 °C. Cell cycle distribution was examined by flow cytometry using a FACScan flow cytometer (Becton Dickinson & Co., San Jose, CA, USA). The relative number of cells in each phase of the cell cycle was analyzed using Modfit program (Verity Software House Inc., Topsham, ME, USA).

Methods
Gene expression profiles and data analysis. Affymetrix GeneChip Human Transcriptome Array 2.0 was used for microarray analysis. Microarray hybridization was performed by the Shanghai Biotechnology Corporation (Shanghai, China) using standard Affymetrix procedures. Raw microarray data were acquired using GCOS1.2 software from Affymetrix. The raw data were preprocessed using robust multiarray analysis (RMA) with a log base 2 (log 2 ) transformation. Gene ontology (GO) analysis was performed to identify the GO terms associated with the differentially expressed mRNA. Tumor xenograft mouse model. Approximately 1 × 10 7 cells were injected subcutaneously into the right dorsal flanks of 4-week-old nude mice (n = 5 per group). Tumor formation was monitored over a 4-week period. Tumor volume was measured weekly and was calculated by the following formula: V = 0.5 × L × W 2 . All animal experiments were conducted in accordance with the institutional standard guidelines of Jiangning Hospital of Nanjing.
Bisulfite sequencing. Genomic DNA was extracted from cells and tissue specimens using the phenol-chloroform method. Bisulfite treatment was performed using a CpGenome TM Universal DNA Modification Kit (Millipore, USA), following the manufacturer's instructions. Modified DNA was amplified, and PCR products were gel-purified and sub-cloned into a pMD19-T vector system (TAKARA, Japan). Ten colonies were sequenced to assess the degree of methylation at each CpG site. The PCR products of the tissue specimens were measured using Sequenom MassARRAY platform (Sequenom, Inc., San Diego, CA). The primers used are listed in Supplementary Table S1.
Chromatin immunoprecipitation assay and ChIP-PCR. The ChIP experiments were performed using an EZ-Magna ChIP G kit (Upstate Biotechnology, Lake Placid, NY, USA) following the manufacturer's instructions. The primers used to amplify the precipitated DNA fragments are listed in Supplementary Table S2. 5-Aza treatment. The DNMT3A overexpression cell line MKN45 was seeded at a density 1 × 10 6 cells/ml in 10-cm dishes. After 16 h of culturing, the cells were treated with the indicated concentration of the DNA methylation inhibitor 5-Aza. Cell culture medium was replaced with freshly added 5-Aza for every 24 h and the total incubation time was 96 h. The cells were harvested for RNA extractions after treatment.
Statistical analysis. The protein expression level of DNMT3A and the mRNA expression levels of p18 INK4C , DNMT1 or DNMT3B in GC patients and the paired adjacent non-tumor tissues were compared using Wilcoxon's test. The correlation between the DNMT3A and p18 INK4C expression levels was analyzed using Spearman's test. The correlation between DNMT3A expression and pathological features was analyzed with the Pearson χ 2 test. Independent Student's t-test was used to compare the results, which are expressed as the mean ± SD between any two preselected groups. A P-value of less than 0.05 was considered statistically significant.